Two Hours a Day, 99th Percentile: How Alpha School Made 2-Sigma Real

MacKenzie Price, founder of Alpha School and 2 Hour Learning, tells Nathan Labenz that there has never been a more exciting time to be five years old. Her own daughters are 17 and 19, and she wishes she could shrink them and start their schooling over. University presidents she has spoken with, she says, are suddenly asking what higher education is for when the half-life of knowledge looks like six months.

That is why Alpha starts teaching kids to use AI as a superpower in kindergarten. Price's framing is simple: models are already good at the black and white of existing knowledge. Humans will have to live on the gray frontier, pioneering new knowledge and new ways of thinking. The school is trying to equip students for that, not for a lecture hall that assumes every fifth grader is in the same place.

Labenz opens with the two-sigma question: could AI finally give every student the kind of one-to-one tutoring that used to be reserved for royalty? Price answers by going back a thousand years. Socrates tutored Plato, Plato tutored Aristotle, Aristotle tutored Alexander, and Alexander had taken over most of the known world by 22. That system was mastery-based and one-to-one. It was also only for the elite.

The industrial revolution, Price argues, forced a cheaper model: a teacher standing in front of a classroom, moving a mixed group of kids through the same curriculum from August to May. For the last 40 years, learning-science papers have claimed students can learn two, five, even ten times faster in a one-to-one mastery setting. Every one of those papers, she says, starts or ends with the same caveat: this does not work in a time-based classroom.

Benjamin Bloom's two-sigma result is the headline version of that claim. Two standard deviations of improvement would, in Labenz's recollection, take an average student into roughly the top 5%. Price is careful not to oversell the original study. People in learning research pick it apart, she notes, depending on the audience and the measurement. What she cares about is the practical version Alpha can run every day.

AI is the first tool, she says, that can measure what a student knows and does not know with enough precision to build a personalized plan, then refuse to move on until the student has mastery. She reaches for a lab-science analogy: when the microscope arrived, biology changed; when the telescope arrived, physics changed. AI, in her telling, is the instrument that can turn learning science from a fuzzy craft that depends on the teacher, the kid, and the home into a practice with clear inputs and clear outputs.

For the 2024–25 school year, Alpha's students learned at 2.3 times the rate statistical models predict. The school uses the NWEA MAP assessment, an adaptive test taken by about 10 million students in U.S. public, private, and homeschool settings. MAP produces a RIT score in math, reading, language, and science, then forecasts how many points a student at a given percentile and age should gain in a year.

If a student at the 85th percentile is projected to gain four RIT points, Alpha's internal bar is eight. Price says the same personalized path works whether a child arrives at the 10th percentile or the 98th: fill holes, then move ahead. The efficiency claim is even starker on a single course. Fifth-grade math, in a traditional school, consumes something like 180 classroom days plus homework, or at least 200 hours. Alpha students finish it in about 20 to 30 hours.

People are incredulous, she says, that this can happen in two hours a day. Her counter is that the time-based classroom is the real joke: only about a third of U.S. students are reading or doing math at grade level. A teacher facing 20 to 30 kids at wildly different levels cannot fill holes and challenge the advanced students at once. The answer, she insists, is not to drop a chatbot on every laptop and ask the teacher to use it for extra lesson planning. It is to get the adult out of content delivery altogether.

The school day is built around what Price calls two hours of productive struggle: focused work on core subjects, often in 25-minute Pomodoro blocks, after which afternoons open up for projects and life skills. Homeschool parents have known for a long time, she says, that six hours in class is not required to crush academics. The scarce resource being returned to kids is time. Motivation, not content, is what the human adults are for.

Headlines call Alpha the school with no teachers and an AI tutor. Students will not describe a cartoon figure that pops up and teaches math. The chat interface, Price says, is the one AI feature they refuse to use. Give a kid a chatbot and it becomes a cheating tool: copy, paste, "what's the answer," "write this essay." Adults may enjoy Socratic back-and-forth with Claude or ChatGPT. Most children will not.

Instead, Alpha teaches K–8 Common Core and high-school AP curriculum through adaptive apps, some bought and some built: ALEKS, Khan Academy, Alpha Math, Alpha Read, Alpha Write, Teach Tales, and others. Students log into Dash, the school's learning platform, and say it is time for math. Dash sends them to the exact lesson in whichever app the system has chosen for that child at that moment. Kids cannot topic-shop out of a hard unit or sit on one question for 15 minutes without the system noticing.

In the background the tutor watches accuracy and speed. If a student skips the explanation after a wrong answer, the system slows them down. If they read the explanation and still fail, it sends them back to a prerequisite, such as multiplication tables before fractions. Price leans on three learning-science ideas: the zone of proximal development, so the work is neither too easy nor too hard; cognitive load theory, so working memory is not overloaded; and spaced repetition, so old material keeps coming back. Vision models now replay the session: the explanation popped up, and the student jumped to the next question in two seconds.

Price founded the first school in 2014, when her daughters were entering first and third grade. Both had arrived at kindergarten already reading chapter books, then sat through lessons on how *A* sounds like "ah." That was a model problem, she says, not a teacher problem: a classroom has to start from the lowest common denominator. Adaptive apps such as Khan Academy and ALEKS already existed, so a third grader could jump to fourth-, fifth-, or sixth-grade work. The apps still let kids bounce around.

Alpha started cataloguing anti-patterns: topic shopping, bailing when questions got hard, doing one item every 15 minutes. A first grader who loved math reached eighth-grade material and needed an adult to read the word problems because he could not read. Alternative schools might have let him skip reading for months. Alpha put up guardrails instead, including 25 minutes of focused work in every core subject every day. Even then, the hybrid era was leaky.

In 2022 language models became usable as measurement and analysis tools. The school began feeding assessments back into the tutor so a test actually changed the next lesson plan, something Price says almost never happens in a traditional building. In the fall of 2022 they made the sharper call: no academic teaching by humans at all. Learning rates moved from about 1.5x to more than 2x, a visible step-change on their multi-year graphs. Edtech bolted onto a time-based classroom for 15 minutes three times a week, she argues, never had a chance.

Alpha has not partnered with a single LLM. The schools are a test bed for whatever is latest, and Price says they burn a lot of money on data analysis and vision models that watch how a child works a problem. She expects that within five years a $1,000-a-year tablet could carry a student's entire academic load. Generative tools are already useful where accuracy is less brittle: Teach Tales writes reading-comprehension stories at a child's Lexile level, with the child as the hero of a plot that weaves in a favorite movie or Little League teammates.

The measurement loop showed its teeth in January 2025. Students take MAP three times a year. That mid-year report came back at the 99th percentile in every grade and every subject except fifth-grade math, which sat at the 93rd. The academic team treated it as a curriculum issue they could debug. In a normal school, Price notes, you cannot tell whether the problem was the teacher, the curriculum, or kids who were not working. Alpha's data is cleaner because the classroom is not in the way.

One finding was product-market fit inside the stack. Math Academy is excellent for advanced students who already like math, and about 25% of Alpha's students thrive on it. Most do better on Alpha Math or similar tools. Depth of knowledge is another axis. During a high-school SAT blitz, one app was strong at DOK levels one and two and weak at three and four; Alpha sent the feedback and the company shipped a new version. Price now employs psychometricians to join learning science to curriculum design, which is why she thinks edtech firms like working with the school.

Labenz asks whether tests and subject boundaries might dissolve. Price does not take the bait. She is not in the camp that wants to replace knowledge with loosely defined project-based learning. The more information a student has to draw on, she says, the better they can think and analogize later. Alpha still drills: she walked into a second-grade morning launch and found kids doing calisthenics while shouting multiplication tables. Kindergarten through second grade already finish academics in about 90 minutes and still leave at the 99th percentile. She would rather compress further, toward one-hour learning, than get "loosey goosey" on knowledge.

The experience is already multimodal. For early reading, students read aloud so the system can score speed, stalls, and pronunciation. Alpha built Fluency Coach and also uses Literably, especially from pre-K through second grade. Price treats third grade as the hinge from learning to read to reading to learn; a weak reader at that point, she says, is in trouble for the rest of school. Phonics, letter formation, typing, matching on a screen, speaking, and listening all run in parallel.

Students with learning differences can benefit without Alpha claiming to rewire anyone. Price is explicit: the school is not changing the neural pathways of a child with dyslexia, which specialized programs try to do. Kids in roughly the bottom 50% of severity for dyslexia or dysgraphia often do well because they can go at their own pace with auditory and visual support. That is personalization, not a clinical intervention.

Handwriting has not been thrown out. Kindergarten and first-grade rooms still have manipulatives, little pegs students can count and lay out so math is visible and tactile. Second graders, asked what independent check is hardest, answered in unison: cursive. Price is not sure how much cursive the new world will need, but she wants tactile work to stay. Online writing apps run alongside paper practice, and lesson plans are allowed to differ by child when one modality works and another does not.

Alpha calls the adults Guides or coaches, and the Guides prefer it. They are not subject-matter experts standing as the sage on the stage. They are the guide on the side, responsible for motivation, mentorship, and emotional support. The school asks older students a pointed question: every adult can name one or two teachers who changed their life; is your Guide that person? Labenz can name several from his own schooling. Price can name two. Both also remember plenty who were not.

Guides are held to three commitments: students love school, they learn twice as fast in two hours, and they develop life skills. Motivation is the lever the humans still own. Kids earn school currency, seating privileges, Chick-fil-A lunch, zoo trips. One Austin student kept staring out the window; his Guide realized he was obsessed with birds, built a poster of species in the local greenbelt, and traded weekly academic goals for 15 minutes of birdwatching. Price's public-school thought experiment is even simpler: finish academics in the morning and give kids back the 3 p.m. activities they already love, football, theater, robotics, debate, dance.

In a traditional school, she argues, failure gets blamed on the child, not smart enough, too lazy, unwilling to turn in homework. At Alpha, if a kid is not thriving, it is the school's fault. After the most recent school year the staff went student by student and asked whether the three commitments were being met. About 15% of Guides come from traditional teaching, especially in the early grades. The rest arrive from startups, pro and college sports, and executive jobs. Starting pay is $100,000. Someone asked whether the schools would have a teachers' union. Price's reply: you will never have the old battle-axe teacher nobody likes who still cannot be moved.

Afternoons are for life skills and for using AI as a superpower rather than a shortcut. High schoolers built an AP U.S. History soundtrack in the spirit of *Hamilton* and used Taylor Swift melodies to memorize AP Art History. Kindergarteners and first graders make books, graphic art, and movies; older students code drones, robots, self-driving cars, and video games. Price's update of the three Rs is the four Cs: critical thinking, collaboration, communication, and creativity.

Flagship Alpha is a high-end private school, with tuition generally between $40,000 and $65,000 depending on the city. Most of that money, Price says, goes into afternoon life-skills workshops, not into a different morning academic product. A gifted-and-talented school, a sports academy, and an esports-focused middle school sit around $25,000. A $15,000 tier is rolling out. Every campus uses the same two-hour personalized platform; the afternoons are what change.

Brownsville, Texas, is the stress test. It is one of the poorest school districts in the country. About half the students are children of SpaceX employees and about half come from the local community, with wide socioeconomic, racial, and English-as-a-second-language diversity. Labenz notes a $10,000 price point there. That campus, Price says, consistently outperforms the others on learning rates. Personalized software, in her phrase, is the great equalizer: it does not care where a child started.

Public adoption is slower. Some states have simply said they are not interested. Arizona approved a virtual charter; Texas is in process for a physical one. Price has talked with the Department of Education and the White House about bringing the model into the public sphere. Her warning to that audience is the same one she gives private schools: do not put a chatbot on a computer and call it a day, and do not tell teachers they are being replaced. The job is supposed to move toward social, emotional, and motivational work. Charter afternoons will not include flying to Poland to train Ukrainian refugees on two-hour learning, but they can include sports, entrepreneurship, and vocational workshops.

The long goal is a billion children. Paths she lists include more private campuses in the U.S. and then abroad, licensing the learning platform, public-school intervention for kids who need remedial or advanced help, and a wave of other innovators copying the pattern. Parents, she finds, already understand the afternoon: financial literacy, entrepreneurship, storytelling, public speaking, leadership, teamwork, grit, and learning to fail. Those are the hours traditional school almost never protects.

Labenz lives in Detroit with a six-year-old and a four-year-old and cannot walk into an Alpha campus. Price's first pointer is Alpha Anywhere, the homeschool program, which had to rebuild the motivation model for families who do not have a Guide in the building. Results, she says, are now good enough to roll it out more broadly. For younger kids, the concrete invitation is Teach Tales: put your child in a choose-your-own-adventure at the right reading level, with their friends and favorite movie inside the plot, and watch them ask to read more.

Public-speaking tools are the other low-friction example. A six-year-old can invent a story, tell it at dinner, then get AI feedback on intonation, pacing, and filler words. An Alpha graduate who just finished her freshman year at Stanford, writing as Austin Scholar, runs a Substack for parents on AI apps and summer projects. Price is also launching futureofeducation.net in June as a free directory of tools. She speaks from experience about after-school scaffolding, science club, writing a book, piano, until the calendar collapses.

The meta-advice is for the adult. In conversations with people in government who want AI-driven schools, Price keeps finding the same block: teachers are not familiar with AI, so they are wary of it. Parents should start in daily life, dinner, work, travel planning, then bring kids in as builders. "Be creators and contributors, not just consumers" is the line she wants families to take away. Labenz, listing the teachers who marked him, had already said the quiet part: if every child had that kind of relationship as the rule rather than the exception, it might matter more than the software. Price's Guides are hired and fired on whether that is true.